AWS Lambda Costs: Memory, Timeouts and Concurrency
An idle Lambda function already costs nothing for compute. The spend sits in memory settings, duration and provisioned concurrency.
What Lambda charges for
Lambda is priced on the number of requests and the duration your code runs, measured in GB-seconds: the memory you allocate multiplied by execution time, rounded up to the nearest millisecond. Provisioned concurrency is a separate charge for the concurrency you configure and the time it is configured. Reserved concurrency, by contrast, incurs no additional charge.
That changes what “turning it off” means. A function that is not invoked has no duration to bill. The money is in how much memory each invocation holds, how long it runs, and any provisioned concurrency kept warm.
What a Lambda schedule really does
Lambda is on ZopNight’s Start/Stop supported list as concurrency throttling. AWS documents that setting a function’s reserved concurrency to 0 stops it processing any events until the limit is removed, using the PutFunctionConcurrency and DeleteFunctionConcurrency API operations. So a schedule is an activity freeze: useful to stop a test function from being invoked by something noisy overnight, not a way to cut the cost of a function that is already quiet.
ZopNight’s Lambda rules
- Lambda Over-Provisioned Memory flags functions whose memory setting looks higher than they need. It is guidance rather than a one-click change, because a memory cut needs load testing first.
- Idle Lambda Function flags functions with no invocations across a 30-day lookback and prices the function’s cost as recoverable on delete.
- Lambda Timeout Too High fires when the timeout is at least five times the observed peak duration. It is a hygiene finding with no claimed saving, because the timeout itself is not billed.
- Lambda Reserved Concurrency Too High is registered but never fires, because reserved concurrency carries no charge.
- Lambda Using Deprecated Runtime covers the posture side.
Findings carry the metric evidence behind them. Scheduling a function needs a read + write connection; a read-only connection still gives you the inventory, cost and every recommendation.
Blast radius before a change
Lambda functions are joined to their triggers in ZopNight’s resource graph. Before acting on a recommendation, the blast radius view shows which connected resources a change would touch, one hop out.
See the Lambda fact sheet, how Lambda throttling schedules work, and serverless cost pitfalls.
Questions we get a lot.
Does scheduling a Lambda function save money?
Only indirectly. An uninvoked function has no duration charge, so a throttle schedule mainly blocks invocations. The larger savings come from memory rightsizing and removing unneeded provisioned concurrency.
What does stopping a function mean on AWS?
AWS documents that setting reserved concurrency to 0 intentionally throttles a function until the limit is removed. ZopNight's Lambda stop is this kind of concurrency throttle.